Google cloud execution

[rsuchecki]

Repset could, in principle, be executed using https://cloud.google.com/life-sciences/

limitations

• not well suited to short-running tasks, this could be addressed by

  • hybrid execution ensuring short-running tasks computed using local executor

• requested memory + CPU combinations must match what is available, e.g.
  Execution failed: creating instance: inserting instance: Invalid value for field 'resource.machineType': 'zones/australia-southeast1-a/machineTypes/custom-8-4783'. Memory should be a multiple of 256MiB, while 4783MiB is requested. Memory size for 8 vCPU instance should be between 7424MiB and 53248MiB, while 4783MiB is requested

  • API calls to obtain available machineTypes & their specs?
  • logic to ensure memory & CPU reservations match?

• ?

Would it be better to go k8s on https://cloud.google.com/kubernetes-engine instead?

To be continued...